Start free trial

Backtest

How our flagged bets did in past seasons

We replayed up to 3 past seasons of NFL, NBA, NHL and MLB player props: every bet our +EV engine would have flagged 60 minutes before the game, at that price, settled against the real result. It's a replay, not real bets, and past seasons don't predict future ones. Most bet types have too few bets to tell skill from luck, and we say so. Game lines, college, Canadian books and some props aren't backtested yet; see below for why.

Licensed US sportsbooks only. Each bet counts once, even when several books flagged it. By bet type and sportsbook ↓

Every flag, and Strategy mode

What every flagged bet returned, and what Strategy mode's filters returned in the same replay.

All flags

−0.1%

return per $100

−10%0+10%

57,433 flagged bets · likely range −1.0% to +0.8% · average flagged EV +3.9%

  • 2025-26:−0.1%on 36,047 bets
  • 2024-25:−0.7%on 17,730 bets
  • 2023-24:+3.2%on 3,656 bets

Strategy mode

Rules set using these same seasons

+4.4%

return per $100

−10%0+10%

8,381 picks · likely range +2.2% to +6.7%

  • 2025-26:+5.8%on 4,434 bets
  • 2024-25:+1.0%on 2,419 bets
  • 2023-24:+6.0%on 1,528 bets

Taken together, our flags roughly broke even: −0.1% against the +3.9% EV they showed, so the EV number alone isn't enough. Strategy mode keeps higher-EV flags and skips the bet types that lost; it returned +4.4% in the same replay. Its rules were set using these same seasons, so its real-world return is likely lower.

Likely range (95%)Actual returnFlagged EVBreak-even

Does a higher EV pay more?

Each bar is what flags in that EV range actually returned. If our EV numbers were exact, each bar would reach the "Flagged EV" mark.

  • 0–3% EV

    −0.8% on 23,112 bets · likely range −2.0% to +0.5%

  • 3–4% EV

    +0.3% on 13,976 bets · likely range −1.3% to +2.0%

  • 4–5% EV

    −1.2% on 8,725 bets · likely range −3.5% to +1.0%

  • 5–7% EV

    +0.8% on 7,477 bets · likely range −1.8% to +3.4%

  • 7–10% EV

    +2.2% on 2,995 bets · likely range −2.8% to +7.1%

  • 10–15% EV

    +7.7% on 913 bets · likely range −3.5% to +18.8%

  • 15%+ EV

    −3.3% on 235 bets · likely range −34.1% to +27.6%

Below 7% EV, returns stayed near zero (−1.2% to +0.8%), and every band's likely range includes 0. Above 7% EV there are only 4,143 bets and the likely ranges are wide, so these bands can't show yet whether a higher EV pays more. Every band returned less than the EV it showed.

Likely range (95%)Actual returnFlagged EVBreak-even

Axis −35% to +35%

Overs vs unders

An over and an under on the same line are opposite bets, so each side is shown on its own and never added together.

SportOverUnderYesNo
All−1.7%24,717+0.8%29,576+4.6%2,705+1.2%435
NFL−1.9%3,554+4.4%3,888+11.1%1,16328 bets
NBA−2.9%†12,516+0.9%13,635+3.4%†1,036+1.6%407
NHL+3.5%1,869−0.8%†924−7.6%506—
MLB−0.7%†6,778−0.5%11,129——

† Strategy mode skips these.

NBA yes props made money only at odds longer than +300, which Strategy mode excludes anyway; inside its −200 to +300 window they lost in both seasons (−5.6% and −15.0%).

Beating the closing line

Closing-line value (CLV) compares the price we flagged with Pinnacle's last price before the game. It's a quicker signal than results, so this shows whether flags that beat the close by more also returned more.

Beat close byBetsReturn · range
Below −2%3,137−2.8%−6.5% to +1.0%
−2% to 0%1,831+1.4%−3.3% to +6.1%
0% to 2%3,425−1.8%−5.3% to +1.7%
2% to 4%10,239+0.5%−1.5% to +2.4%
4% to 7%5,757+1.4%−1.4% to +4.2%
7% or more2,065+5.0%−1.7% to +11.8%
No Pinnacle close*30,979−0.5%−1.7% to +0.6%

Every band's likely range includes 0, so at licensed books this replay can't show the link yet.

* Pinnacle doesn't post every prop, so these flags have nothing to compare against.

How many bet types have a verdict

A bet type is one sport, market and side, like NBA points unders. Most have too few bets to tell skill from luck; a 5% edge needs about 1,500 bets to show.

  • 1 held up every season
  • 3 beat the closing line
  • 4 lost every season
  • 61 no clear result yet
  • 11 too few bets

A verdict needs 100+ bets and results from at least two seasons. With this many bet types, one or two will hold up or lose every season by luck alone. Members can see every bet type below.

What we haven't tested yet, and why

The replay covers player props in the four pro leagues. Everything else is listed here with the reason, so you know which bets have no history behind them.

SportPropsMLSpreadTotal
NFL14 of 29testedNot yetNot yetNot yet
NBA9 of 13testedNot yetNot yetNot yet
NHL4 of 8testedNot yetNot yetNot yet
MLB10 of 13testedNot yetNot yetNot yet
NCAAFNot yetNot yetNot yetNot yet
NCAABNot yetNot yetNot yetNot yet
Moneyline, spread and total (NFL, NBA, NHL and MLB)
Not tested yet. The replay fetched player props only.
NCAAF and NCAAB
Not tested yet. The replay covered the four pro leagues only; college props also have no Pinnacle price to measure against, and no feed to settle them.
Props we offer without enough history
  • NFL: 15 of 29 markets4 not in the replay6 never flagged3 flagged too few times2 short of a verdict
  • NBA: 4 of 13 markets1 not in the replay1 never flagged2 short of a verdict
  • NHL: 4 of 8 markets2 not in the replay1 flagged too few times1 short of a verdict
  • MLB: 3 of 13 markets1 not in the replay2 never flagged

Not in the replay: a standard box score can't settle it (for example, who scores first). Never flagged: too few books post it, and none of them sharp, to set a fair price. Too few or short of a verdict: flagged, but under the 100 bets a verdict needs.

Sportsbooks
11 US books tested. Canadian books not yet. The historical prices we used cover US books only.
Alternate lines
Alternate lines aren't in this replay. A separate pilot tests them for the alternate-line model.
Arbitrage and middles
Not backtested. An arbitrage's return depends on getting both prices before they move, not on the result, so a replay can't measure the real risk. Middles aren't tested yet.
Pick'em
Not backtested. Pick'em apps pay fixed multipliers instead of odds, so they need a test of their own.

By bet type and sportsbook

Every market and side we replayed, grouped by verdict, and what each sportsbook's flags returned.

See it by prop type and sportsbook

Members can break these results down by market and side for each sport, see each sportsbook's return on the flags it carried, and jump from any row to today's live bets like it.

Start your 3-day free trialLog in

Then US$29.99/month, every tool included. Cancel anytime. 21+ in the US · 19+ in Ontario.

How to read this

Return per $100
The average won or lost per $100 staked. +2.0% means $2 won per $100, on average.
Likely range
Where the long-run return probably sits (a 95% range). Few bets give a wide range. If it crosses 0, the result could be luck.
Flagged EV
The edge our engine showed when it flagged the bet.
Verdicts
A verdict needs 100+ bets and results from at least two seasons. Smaller bet types show their bet count only.
Luck
A 5% edge needs about 1,500 bets to tell from luck. Across dozens of bet types, a few will look good or bad by chance alone.
The replay's limits
Prices were taken 60 minutes before each game. Yours may differ, books can limit accounts that win, and past seasons don't predict future ones.

The Smart Lines is strictly an analytics tool. We don't take, place or facilitate bets, and we don't condone or encourage gambling. Use the information at your own judgment and risk. If you or someone you know has a gambling problem, call 1-800-MY-RESET in the US or ConnexOntario at 1-866-531-2600 in Ontario, or see our responsible gambling page. 21+ in the US · 19+ in Ontario.Updated Oct 13, 2026